Development and Application of RemoteSensing of Longwave Gooling from the

NOAA Polar Orbiting SatellitesRobert Ellingson, David Yanuk, Arnold Gruber, and A.f. Miller

AbstractSatellite data have provided considerable information on therudiation budget at the top of the EaXh-atmosphere system,However, in order to maximize the usefulness of these observations, it is necessary to know how the radiative heatingand cooling are distributed within the atmosphere and be-tween the Earth's surface and the atmosphere. A techniquehas been developed to use rcdiance data from the High Res-olution Infrared Sounder (uns) instrument flown on NzAAoperational satellites to obtain estimates of the profile oflongwave atmospheric cooling (tc) and the atmosphericemission to the Earth's surface (downward longwave radia-tion, DLRJ. Bfiefly, the nta and tc are estimated from umsradiance obseruations using regression techniques on radia-tive tansfer calculations. The technique rcquires the spec-tral rcdiance data from HIRS and the vertical distribution ofcloud amount and cloud-base and cloud-top heights. Cloudinformation is not generally available concomitantly withthe wns radiances, and the initial effort has focused on thedevelopment of clear sky models. Radiative cooling is calcu-lated for four layers: surface to 700 mb, 700 to 500 mb, 500to 240 mb, and 240 to 10 mb.

Initially, a month-long data set was produced - 15 De-cember 1990 through 15 lanuary 1991 - for study and tech-nique evaluation. Calculations were global on a 2.5o by 2.5oIatitude-longitude grid. Monthly averages and five-day run-ning means were produced. Comparisons were made to theNational Meteorological Center (NMc) medium range forecast(MnF) model fields of tc and nta. The agreement was generally within values expected from comparisons of calcula-tions from the dffirent models, especially for zonallyaveraged quantities. There were, however, significant differ-ences over specific geographical ateas (e.g., Africa and Aus-tralia). Analysis of these differences indicated whereimprovements were needed in the HIRS and the MRF tech-niques, resulting in an improved uns model for estimatingclear sky DLn and LC.

The clear-sky algorithms for the LC and outgoing long-wave radiation at the top of the atmosphere have been im-plemented as an experimental quasi-operational system forfurther evaluation. Twelve months of data (June 1992through May 1993) have been processed to date, and theavailabilitv of the data were announced to the international

R. Ellingson and D. Yanuk are with the Department of Mete-orology, University of Maryland, College Park, MD 20742.

A. Gruber and A.J. Miller are with NOAAAIESDIS, Washing-ton, DC 20233.

PE&RS

climate community for use and evaluation beginning in Jan'uary 1993.

lntroductionRadiative processes in the Earth-atmosphere system are es-sential for maintenance of the atmospheric general circula-tion. Absorbed solar radiation is the principal source ofenergy, whereas energy lost to space through terrestrial ra-diative processes is the ultimate energy sink. Most of ourinformation on the time varying geographical distributionsof the radiation source and sink at the top of the Earth-at-mosphere system has come from Earth orbiting satell i tes.The better known experiments that measured the planetaryshort- and Iongwave radiation budget were Nimbus 6 and 7nns and the Earth Radiation Budget Experiment (ERBE)(Barkstrom ef a/., 1989), Operational estimates of the top ofthe atmosphere radiation budget are provided by NoAafrom the polar orbiting satell i te data (e.g., Gruber and Win-s t o n , 1 9 7 B J .

These data sets have limited utility because they onlyyield an estimate of the radiative cooling for the entire Earth-atmosphere system, In order to gain maximum utility fromthese data in forecasting and climate diagnostics, it is neces-sary to know how the radiative cooling is distributed be-tween the atmosphere and the surface. This has been clearlydemonstrated by Slingo and Slingo (1988; 1991) and Ell ing-son and Campana (1987). Such information is of interest aswell to all "three streams" of the World Climate ResearchProgramme (wcRP) (WCP-115, 1986): extended range fore-casting, time scale of 1 to 2 months; interannual variation,time scale of several years; and time scales of several years toseveral decades. Because the atmosphere is mostly transpar-ent to shortwave radiation, the maior portion of the atmos-pheric radiative heating (cooling) is accomplished byIongwave radiation. (The terms longwave and shortwave areused herein to denote radiation of wavelengths greater thanor less than 3 micrometres, respectively,)

At the start of this project there was no operationalproduct available to scientists with which to study the rela-iionship of this cooling with climate and long-range weatherforecasts, although the need for knowledge of the distribu-tion of radiative cooling had been established. In 1987 we

Photogrammetric Engineering & Remote Sensing,Vol. 60, No. 3, March 1994, pp. 307-316.

oo99-77 721 S 4/6003-3 07$03. 0 0/0@1994 American Society for Photogrammetry

and Remote Sensing

proposed to NASA to rectify this gap by developing an opera-tional methodology for providing estimates of longwave cool-ing on a global scale from satellite observations. Theobjective of the proposed project were to (1) develop andevaluate the methodoloev for determinins the vertical dis

'oposed prolect were to [1J develop andodology for determining the vertical distri-

bution of iongwave cooling from the NOAA High ResolutionInfrared Sounder (urRs) onboard the ruoaa operational satel-l i tes, and (2) implement the developed procedure into a pilotoperation for test and evaluation as input into an operationalforecasting model and for use as a climate diagnostic tool.This paper summarizes our activities to meet those objec-tives.

Structure of the Project TeamThe project consists of teams at the University of Maryland(uuo), headed by Ellingson; NoAAAIESDIS, headed by Gruber;and Non A.hlc, headed by Miller. The idea for the projectarose from discussions between Ellingson and Gruber at theNOAA Cooperative Institute for Climate Studies (CrCS) at UMDabout using or extending the NOAA operational radiationbudget product to estimate vertical profiles of longwave ra-diative cooling rates. Similar discussions had produced atechnique for estimating the outgoing longwave radiation atthe top of the atmosphere (orn) from the operational soun-ders (HIRs) (Ell ingson ef 01., 1989a), and Ell ingson's experi-ence with radiative transfer theory led him to believe thatsuch a technique was possible with the HIRS data. Further-more, both Gruber and Miller had extensive exnerience inthe dlvelopment and analysis of data from NOnR's opera-tional programs and in the dissemination of data to [he inter-national meteoroloqical communitv.

The team at uiln had the primary responsibilities for de-veloping the inference techniques, developing the technologyfor displaying the data, interpreting the differences betweenestimated and calculated fields, and distributine the data in aquasi-operational mode. This activity was suppirted by thefull-time assistance of a faculty research assistant, two full-time graduate students, and one-fifth of Ellingson's time. Thecomputer resources included a dedicated Apollo worksta-tion, unlimited access to the UMD Meteorology Departmentslocal area network of nrc and Apollo workstations, access toUMD's computational facilities, and access to the Internet.

The NOaaAteSDIS activitv provided access to the ooera-tional satellite data stream, pro^vided assistance with theinterpretation of the observed and inferred satellite data, andprovided guidance with streamlining the technique to opera-tional status. This activity was assisted for one year by aUMD faculty research assistant along rvith about 10 percent ofGruber's time. The project provided a workstation which wasconnected with the NOAA operational network. The worksta-tion received the HIRS data, performed initial data reduction,and forwarded the data to UMD for further processing.

The NoAAAIMC participants provided operational wuCproducts for use in testing the inference schemes, providedguidance on the interpretation of the NMc and inferenceproducts, and provided guidance on the structuring of thedata products for the potential user community, This activ-ity was accomplished by input from several NMc scientistswhich amounted to about 10 percent of a man vear peryear.

Cooling Rate And Surface Flux Estimation TechniquesThe vertical distribution of temperature, water vapor, andcloudiness are the primary modulators of the thermal radia-

tion emitted to space and to the surface. As such, measure-ments of radiance emitted bv the Earth-atmosphere svstem tospace have been used to estimate profiles of the atmospherictemperature and humidity since the advent of man-made,Earth-orbiting satellites. In principle, the temperature andwater vapor profiles inferred from routine satellite observa-tions, when combined with cloud estimates. could be usedin radiative transfer models to calculate radiation energybudget quantities such as vertical profiles of upward anddownward Iongwave fluxes and cooling rates. This type ofapproach has been done routinely since the mid 19801 at themajor medium range (3 to 5 day) weather forecast centers inEurope (the European Center for Medium Range WeatherForecasts, ECM\IT) and the U.S. (the National MeteorologicalCenter, NMC) at the start and during the daily forecast cyclesusing radiative transfer theory and the initial and forecastvariables. Such calculations require extensive amounts ofcomputer time, and errors in the forecast model assimilationand satellite inferred temperature and humiditv fields mavlead to undesirable ..rots i.t the calculated flux and coolinerate fields.

An alternative approach is to use the radiance observa-tions to directly infer the fluxes and cooling rates withoutperforming tha intervening radiative transdr model calcula-tions. This approach is physically reasonable because theoutgoing longwave radiance measured in space is part of thetotal outgoing radiation flux at the top of the atmosphere(oLR), and it is an integral measure of the atmospheric cool-ing to space for a given spectral interval. Rodgers and Wal-shaw (1966) showed that "cooling to space" ii the majorcontributor to the total longwave cooling (LC) at any level inthe atmosphere. Theoretical studies by iiou and Xue (19S6)have shown that it should be possible to estimate verticalprofiles of t c from radiance observations, More recently, EII-ingson ef o,l. (198ga; 1993) have shown that a Iinear combi-nation of radiance observations from the HIRS on the NOAAoperational satellites may be combined to produce estimatesof the oLR at the top of the atmosphere that rival the accu-racy of those observed by ennr, and Lee and Ellingson (1990)showed that similar procedures could be used to obtain anaccurate estimate of the downward radiation flux at theEarth's surface (oln), Because the Ellingson et al. (19S9a)technique offered the possibility of rapid, and potentially ac-curate, estimates for global scale analyses, we decided topursue its extension to the derivation of vertical profiles ofIongwave cooling (Lc).

Briefly, the vertical profile of Lc is estimated from HIRSradiance observations using regression techniques based onradiative transfer model calculations. The cooftng rate for thek+h Iayer is estimated in a manner similar to the way itwould be calculated in a radiative transfer model, namely

LCo : 2 a*i N,r * Aj > bHr N.t (1)

where the a's and b's are coefficients determined fromregression analysis of model calculations, the N.1's arc cloud-cleared radiances obtained from the operational analvsis ofthe measured radiances, andA is the iffective cloudiractionat-level j (multiple, randomly overlapped, cloud layers areallowed in the analysis). The first term on the right handside of Equation 1 is the clear-sky cooling rate and the sec-ond term is the difference between the completelv overcastand the completely clear terms. Note that the cloud-clearedradiances contain information on the vertical distribution ofatmospheric variables other than clouds.

PE&RS

The spectral intervals and the regression coefficients forEquation i were determined with a siepwise reqressionanalysis of calculations frorn a theoretiial radialion modelusing 1600 soundings as input data. The necessary intensity,flux, cooling rate, and radiance data were calculated with aversion of the radiation model described by Ellingson andSerafino (1984). Briefly, the model uses the Malkmus (1967)band model fitted to the 1986 AFGL spectral line data forHrO, CO, 03, CH4, and NrO in the spectral region between 0and 3000 cm-1. (i.e., all wavelengths less than 3.3 mm). Thewater vapor continuum is included using a modification ofthe parameterization given by Roberts et al. (7s76) so that itmore nearly reproduces the data of Burch and AIt (1984).Variations of the band model parameters along the atmos-pheric path are included using the Curtis-Godson approxi-mation as described by Rodgers and Walshaw (1966).

The atmosphere is assumed to be axi-symmetric, andplane parallel out to 80", beyond which spherical symmetryis assumed. Specific intensity is calculated at seven angles ineach of 140 spectral intervals ranging in width from 5 to 40cm-1. The necessary integrations over altitude and zenith an-gle are performed with trapezoidal and four-point Gaussianquadratures, respectively. The application to the 19 HIRSchannels was done using NOAA supplied, Iaboratory mea-sured values of the instrument response functions.

Cooling rates were computed in each of four layers (t000to 7o0,700 to 500, 500 to 240, and 240 to 10 mb), The effectsof clouds were included by computing cooling rates in theindividual layers with complete (black) cloud cover below,within or above the given layer with cloud thicknesses rang-ing upwards from 1 mb. For a given layer and cloud basepressure, regression coefficients were calculated as a func-tion of cloud layer top. Cloud bases and tops were alignedwith model levels to span the range of possible cloud loca-tions and thicknesses. It should be noted that the atmos-pheric water vapor profiles were not modified to yield 100percent relative humidity in the regions of assumed cloudi-ness. Actual cloud emisiivities may included by multiplyingthe cloud fraction by the appropriate emissivity. In opera-tional use, the precalculated regression coefficients may beinterpolated to match the observed cloud locations.

The 1600 soundings used in the radiative transfer calcu-iations were compiled by Phil l ips ef o1. (19s8). Briefly, eachsounding includes temperature values at 65 different pres-sure levels from 0.1 to 1000 mb and mixing ratios of HrOand O. in the corresponding 64 layers, The soundings werecomDiled from radiosonde ascents from land and ocean sta-tioni between 30"S and 60"N latitude, and the soundings areequally divided between tropical (30"S to 30"N) and midlati-tude (summer and winter) conditions. The O" data were cho-sen to be climatologically consistent with the temperatureprofiles, and the stratospheric HrO mixing ratio is assumedto be 3 ppmm. The surface skin and air temperatures wereassumed equal.

The regression analyses used to develop the oLR, LC, andDLR estimation techniques were performed with a least-squares, stepwise, backward glance technique of the type de-scribed by Efroymson (1960), A variety of subroutines fromthe S* (see SPLUS, 1992) and IMSL iibraries (see IMSL, 1984)were used in the analysis, and 0.05 was specified for the sig-nificance level to add or delete variables. The variances ofthe coefficients were estimated following the techniques il-iustrated by Anderson and Bancroft (1952J,

The determination of the form and coefficients of the

TISLE 1. Suvtr,rlnv INFoRMATIoN Colcenltruo rxe Resulrs or rne Recnesstor'tANALysrs roR Oerntt'tttlc Lotcweve Cooutrlc RATES FRoNI HIRS Rnotrrcr

Ossenvelo^ls

HIRS ChannelNumber and

Atmospheric CentralLayer (mb) Wavelength

Level of RMSPrincipal Peak cooling

Absorbing Energy rate errorConstituents Contribution ('C/day)

1000-700

700-500

500-240

240-70

6 (13 .70pm)I (11 .10pm)

77 (7 .30pm)

12 (6 .70pm)

I (9.90pmJ11 (7 .30pm)

2 (14.70p.m)

coralzo 800Window SurfaceH"O 7OO

H,O 500 o.28

O.AlrO 25lSurface o.28HrO 7oo

anv v 2 6 0 0 . 7 2

regression equation consumed a large fraction of our researchefforts during the first 24 months of the project. Our analysisshows that linear combinations of HtRs radiances yield un-certainties in individual estimates of LC on the order of 10 to1.5 percent relative to the mean (about 0.3'C/day nvs), re-spectively, for homogeneous clear or (black) cloudy scenes.The largest errors occur in the two Iowest layers where thetechnique tends to underestimate the actual cooling for Iargecolumn water vapor amounts (i,e., tropical regions)' Thespectral intervals chosen for the clear-sky LC analysis arelisted in Table 1 along with information concerning the HIRSchannels and the RMS estimation errors. The coefficients maybe obtained from Ellingson (Internet email-bobe@atmos.umd.edu).

It is often difficult to attribute physical significance topredictors selected in regression analysis of geophysical databecause of the intercorrelations between the variables. How-ever, the spectral intervals chosen by the regression analysiscoincide quite closely with our knowledge of the physicsgoverning the cooling in the various layers. For example,-ooling in the 1000 to 700 mb layer occurs mainly in theportion of the spectrum from B to 12 pm and is controlledlargely by the surface temperature and the total amount ofwater within and above the layer. Likewise, the variableschosen for the 700- to 500- and 240- to 10-mb layers reflectthe dominance of "cooling to space" by these layers. Coolingin the 700- to 500-mb layer is controlled primarily by watervapor emission in the 2O- to 25-p'm region, and this is highlycorrelated with emission in the 6.7-pm region. In the strato-sphere, cooling to space by the 1s-fm CO, band dominatesthe exchange process. Cooling in the layer from 500 to 240mb is controlled by water vapor emission in the 20- to 50-pm region, but the HIRS channeis are not very sensitive tothis emission. Nevertheless, the channels chosen by theregression analysis reflect the intervals that best explain thevariance in this region,

The cooling rate errors are sufficient for many climato-logical studies, particularly because uncertainties in mea-sured or inferred temperature and water vapor profiles canIead to errors of this magnitude in radiation model calcula-tions (Ell ingson and Gille, 1978). Therefore, we decided touse this formulation for all of our LC analyses. Nevertheless,it is quite likely that the instantaneous errors might be re-duced by a more detailed analysis to account for the insensi-tivity to high concentrations of water vapor (see below)'

It strouta be noted, however, that due to the non-random

0 .31

nature of atmospheric profiles, this type of technique maylead to systematic errors. In mid-latitude regions, where iheatmospheric temperature and moisture proflles change fre-quently, the uncertainties over a monthl time period-largelycancel. ilowever, systematic errors will likely persist in [roi-ical regions (see Ellingson et 01., 1989b). Neveitheless, thelarge scale horizontal patterns should be largely correct. Wehave not looked closely at the magnitude of*the signs of theerrors in successive layers, which is related to the change inthe stability of the atmosphere. The technique could belm-prgv.ed in this-regard !V only choosing combinations of spec-tral intervals that confine the uncertainty of the firstderivative to within particular limits. Modifications to re-du-ce the magnitude of the estimation errors may require adifferent approach to the regression analysis, *hich-is Ieft forfuture research.

Model Calibration and ValidationFlux and radiance calculations from the radiation modelused to develop the oLR, ntn, and LC estimation techniqueshave beel compared with observations from a variety oisources (satellite radiance data, Ellingson and Gille (rsza);aircraft pyrgeometer data, Ellingson and Serafino (19Sa);ERBE oLR data, Ellingson et al. (1994)), and the calculationshave been found to agree to within the rather large uncer-tainties of the observations. Furthermore, the longwave ra-diative transfer model was tested in the framewoik of theIntercomparison of Radiation Codes in Climate Models(rcnccv) (Ellingson ef 01., 1991). As such, it is calibratedagainst international standards. The study bv Ellinsson ef o1.(1994) shows individual oLR estimates to-agree witir nnsndata to within 5 Wm, RMS for all cloud and surface types.However, the LC and DLR estimation techniques have nbtbeen calibrated with observations directly, ilthough that isbeing don-e as patrt of our participation in the NASA spon-sored surface radiation budget algorithm intercompaiisonstudy, the FiRE Cirrus II Experiment (Cox et o1., 19-BZ), andthe DOE sponsored Atmospheric Radiation MeasurementsProgram (DOE, 1990).

The application of any model-based inference scheme toobserved radiance data requires consistency between the ob-served and model-calculated radiances. That is, although thefluxes calculated by the model may be in excellent agrJe-ment with observations, the radiances in some nauow sDec-tral intervals may show Iarge differences with observed data,Such disagreements will Iead to large errors in the inferred_quantities when the model-based coefficients are applied tothe observations,

_ To_guard against this possibility, we used clear-sky ra-diosonde observations collocated with NOAA-10 HIRS over-fliglrts in an attempt to determine the magnitude of the errorsin the radiance calculations. On the basigof 2667 calcula-tions, we found the model to have mainly spectrally varyingsystematic differences with the observations in HrRs chahnels1 to 7 (i.e., the 15 mm CO, band). Furthermore, the analysisshowed that some of the regression estimates were particu-larly sensitive to small variations in some spectral intervals.Therefore, weve modified the regression analysis to eliminatespectral intervals which appeared to yield gross errors in theLc fields. The systematic differences are subtracted from theobserved radiances when the estimation equations are ap-plied to observed data. Although this is tantamount to tuningthe analysis procedure, this step is necessary in order to in-

-

sure the stability of the retrieval technique. Such modifica-

3ro

tions will become less important as radiation models areimproved. Overall, on the basis of our comoarisons withERBE, our limited comparisons with aircraff and surface fluxdata,-our intercomparisons with Iine-by-line calculations,and from the regreision analysis, we eitimate the RMS accu-racy of individual clear-sky OLR, DLR, and tc estimates to beabout 5 Wm,, 15 Wm,, and 0.35"C/day (outside of tropicalregions), respectively.

As noted above, the DLR and LC estimation techniquesrequire spectral radiance data from the HIRS and the verticaldistribution of cloud amount and cloud-base and cloud-topheights, the latter of which are not available on an oDera-tional basis. Furthermore, the radiation model and tlie infer-ence schemes have not been calibrated adequately for cloudyconditions. Nevertheless, we have determined the coeffi-cients necessary for the inference techniques for cloudy con-ditions and we are testing them for potenlial operationilimplementation using cloud distributions from U.S, AirFor_ce RtwEPH analyses and climatological cloud locationsand joint probability information. The uncertainties in DLRcalculations due to uncertainties in the cloud base altitudeare estimated by Frouin et o1. (19S8) to fall between 4 and gWm2.km-1 for tropical and subarctic winter conditions, re-spectively, for 100 percent cloud cover. These estimates arelinear in cloud fraction. The uncertainties for coolinq ratesare not-as easily quantified as they depend upon the'posi-tions of the clouds relative to the layer boundaries. Eirors ofseveral degrees per day are possible-in individual LC esti-mates-by misplacing the position of the cloud top.- Although we have developed the technique to accountfor all types of cloud conditions, the results Jho*n in the re-mainder of the paper are restricted to clear-sky conditionsdue to the lack of a detailed cloud analyses af the time of theradiance observations, Effects of cloudiwill be covered insubsequent publications.

Preliminary 0perational TestingA. 1o-nth long operations study was carried out for the pe-riod from 15 December 1990 through 15 January 1991 for thepurpose-of testing the throughput of the technique in an op-erational setting and for qroviding a data set foiinitial appli-cation by NMC. In particular, the initial application consislsof a determination of the agreement betw^een NMC calculatedand satellite determined fields of clear-sky oLR, DLR, and tC.Clear-sky conditions were chosen because of the lack of anoperational cloud product for each procedure. Therefore,comparisons ofthe global fields of radiation budget quan-tities produced by our technique with the numerical modelsyield estimates of deficiencies in the initialized fields as wellas in the inference technioue.

During this time the iloud-cleared radiances from eachNOAA-1o orbit were obtained from the NESDIS ooerationaldata stream along with clear-sky OLR, DLR, and-lc calcula-tions from the initializations of the mlc Medium Range Fore-cast (Mnr) model. The data from each source wereinterpolated_to a commo_n 2.5' latitude by 2.8. longitude gridto produce the daily and monthly averaged clear-sky radi-ance, oLR, DLR, and tc fields,

Certain characteristics of the data stream are importantto the interpretation of the results, First, the NoAA-16 satelliteis a sun-synchrono rs polar orbiter with an equator crossingtime of approximately 0730 and 1930 local time, The initia-li-zations of the vnr are performed globally at 00, 06, 72, and18 UTc with observations obtained at or interpolated to those

PE&RS

times. The initialization data include satellite inferred tem-peratures, but the water vapor distribution is based-upon themodel spin-up parameterizations. Thus, the tvtRt radiationfields include the effects of a more highly resolved diurnalcycle as compared to the HIRS estimates. Furthermore, theglobal inferred fields are not synoptic (i.e., not determinedeverywhere at the same time).

Second, the individual mRS cloud-cleared radiancesare representative of approximately a 250 km2 area onEarth, but these are not spatially continuous, becausecloud-cleared radiances are not determined by NOAA inovercast regions, However, over the course of the month,about 97 percent of the gridded boxes had observations.We approximated the missing data on a daily basis by in-terpolating either linearly in longitude or bi-linearly in lat-itude and longitude, depending upon the density of clear-sky radiance field.

It should also be noted that the NMC MRF radiationmodel does not include the same parameterizations of theabsorption properties of atmospheric gases used to developthe inference schemes. The longwave radiation model usedin the MRF follows the work of Fels and Schwarzkopf (1975)with modifications as described by Schwarzkopf and Fels(1991). The model has been compared extensively with line-by-line (lnl) calculations and with our model as part oftinccv (Ellingson et a/., 1991). It should be remembered thatthere is no absolute standard for comparisons, but LBL calcu-lations most accurately account for the known physics. Thepublished IcRccM results show the Fels-Schwarzkopf modeland Lgl calculations to agree to the order of 1 and 3 Wm'zfor the oLR and DLR, respectively, and to about O.1"C/day forone kilometre average Lcs in the troposphere. However, theIcRccM calculations did not include the effects of CHo andNrO. The results from ICRccM combined with the MRF-miss-ing effects of CHr and NzO indicate that the unr modelyields higher (lower) oLR (DLR) values relative to our modelof about 6 (12) Wm'z when it is applied to the same tempera-ture and water vapor profile. We estimate the model coolingrates to agree to within about t 10 percent for the layersused in this study.

Because the data used for the MRF come from the modelinitialization rather than from the satellite observations, onecan not rely solely on the above numbers to estimate the ac-curacies and magnitudes of the expected agreement betweenthe vnr and satellite inferred values of oLR, nlR, and t cs.For the DLR, the review by Schmetz (1989) gives differencesbetween monthly averaged DLR observations and calculationsusing data from the TIRos Operational Vertical Sounder(rovs) on the order of 10 Wm', Using LowrRANz (Kneizys etaI., 1988), we have calculated the effects of uncertainties inthe Tovs retrieval to be on the order of t0 and 20 Wm'z fortemperature and water vapor, respectively, for individualclear soundings, For the oLR, these effects are of the order of5 Wm',

We suspect the errors in the MnF temperature initializa-tions to yield similar nLR and oLR uncertainties as notedabove over oceanic regions, and somewhat Iarger errors thanour calculations due to improper specification of the watervapor field in data sparse regions (the tropical oceans andAfrica). Because the temperature of the Earth's surface is cal-culated by the vnr over the continents from energy balanceconsiderations as part of the models initialization, we antici-pated potentially larger discrepancies in these regions, par-ticularly at times of maximum heating or cooling. For

PE&RS

cooling rates, we estimate the uncertainties in the MRF cool-ing ratis on individual profiles to. be at lee,ct on the order ofldpercent (see Ellingson and Gille, 1978) because the uncer-tainties in the unr pioducts are as large as or larger than theerrors in radiosonde data. The random and/or systematic na-ture of these differences follows our previous discussion.

Overall, for monthly averaged data, we expect the zon-ally averaged differences to be similar to the differences be-tween model calculations using the same profiles, becausewe will have averaged over effects of both systematic andrandom errors in the regression estimates as well as the ini-tialized meteorological Iields (assuming both lositive andnegative systematic emors ate uniformly distributed around alatitude ciicle). However, there are likely to be large differ-ences at any given location due to persistent systematic er-rors in both tle MRF and satellite inferred analyses.

Plates 1 and 2 show the infened monthly averaged,clear-sky global distributions of the oLR, DLR, and the 700- tosoo-mb and 500- to 240-mb LCs and their differences relativeto the MRF results for the study period, respectively ' Zonallyaveraged inferred and trlRr calculated values of the lc andoLRlDiR are displayed in Figures t arLd 2, respectively, Ingeneral, the clear-sky global fields are zonal in nature (i,e.,[he major changes are in the north-south direction) due tothe zonal nature of the large-scale distributions of tempera-ture and water vapor. The larger diurnal variation of temp,er-ature near the surlace over continental regions is responsiblefor most of the east-west variation of the DLR and ot R,whereas east-west water vapor variations are the largest mod-ulators of the zonal character of the layer cooling rates,

Overall, the comparisons of the zonal averaged nlns in-ferred with MRF calculated quantities show relatively goodagreement with the oLR and DLR agreeing-to-within the ex-p-ect.d differences at most latitudes, Similarly, the large-scaleoatterns of the clear-skv elobal distributions of all of thequantities are in generally good agreement-' These featuresadd further credibility to the inference techniques' However,there are large differences in the fields of all inferred qua-n-tities in some geographical locations, particularly over Africaand Australia lnd some tropical oceanic regions, that cannotbe explained by differences in the parameterizations of theradiation physics,

> 2

! rq 0

'9 Ib0. E 3O o

t000 - 700 mb

_HIRS_-_._._ rWtF

500 - 240 mb

"/ -\-_

45 0 -46 -90

Latitude (Degrees)

Figure 1. Comparison of zonally averaged clear-sky coolingrates from the HIRS and the MRF for the period from 15December 1990 through 15 January 1991.

r F. I'9 'o

3 r l

HIRS OLR HIRS DLR

r:ii{iJrw

275 300 t25 175 22ft 275 325 375 425

.35 -26 -15 -5 5 15 25 35 Wlmz

Plate !, Global distributions of the HlRs estirnated oLR, DLR, MRF - HIRS oLR and MRF - HtRs DLR for the period from15 December 1990 through 15 January 1991. Flux units are Wm2,

MRF - HIRS OLR MRF - HIRS DLR

m - . . , l + ' r n q \ f f i-126 -roo -75 -60 -25 0 26 60

It is the analyses of the differences that wil l lead toimprovements in the uRr initialization and the HIRS infer-ence techniques. For example, we were particularly con-cerned with the OLR differences over the dessert regions,because many of the differences were almost an order ofmagnitude greater than the rms differences seen in pre-vious comparisons with the ERBE. We were concerned thatthey might be do to a combination of regression errors, theasynoptic nature of the satell i te data, or perhaps to Iargetemperature gradients near the surface. These differencesmight also be responsible for some of the differences seenin the LC and orn fields.

To test for these possibilities, the MIIF initializationfields for one day were used as input to our detailed radia-tion model to calculate the global fields of oLR, DLR, Lc, andthe HIRS radiances at each of the synoptic times. The calcu-lated radiances were then used as input to the inferenceschemes to provide regression estimates of the same fields atthe synoptic times. A variety of comparative studies werethen performed to test the inference schemes. The results ofthose comparisons show that only about 25 percent of theOLR and LC differences seen over the continental regions maybe explained by the asynoptic sampling and/or the largenear-surface temperature gradients. We now suspect that thedifferences are due to errors in the upr initialization in theseregions, These results have been shared with NMc, and thev

312

are examining tle problem to see if it continues during theoperations test discussed below. Additional studies of thedifferences in the LC fields are underway.

As seen from Plate 1, there are differences in the DLRfields that amount to as much as 100 Wmr. The largest dif-ferences occur in regions of high terrain which were nottaken into account in the results shown. Nevertheless. asidefro-m the orographic effects, there are many regions where thedifferences are substantially greater than the purpone1.5{m, RMS accuracy of the DLR technique. Our-analysis ofthe detailed model calculations discussed above founi thatthe DLR model which is linear in radiance has serious short-comings when there is either high or low column water va-por amounts. This led one of our students, Hai-Tien Lee, tocompletely reanalyze the ol,R technique and the radiationmodel calculations,

_ In a recently completed dissertation, Lee (1993) hasshown that the clear or overcast DLR may be accuratelv in-ferred from the HIRS radiances with an equation of the form

DLR = oTa (N.ro) e(N"r, N.u, N.ro, N.rr) (2)

where o is the Stefan-Boltzmann constant, l" is brightnesstemperature corresponding to N"ro, and the emissi,Iity func-tion e is fitted to non-linear functions of the indicated clear-column HIRS radiances. The functional form of e and the

PE&RS

HIRS LC HIRS LC

iLllii'3riffi

2 2.6.r\sm2 2.6

MRF - HIRS LC500 - 240 mb

ff ir i l i :S\W

..75 -.5 -.25 0 .26 .6 .76 1|+,*,rffi

.cld.ay -.76 -.6 -.26 0 .26 .6 .ltt 1

Plate 2. As in Plate 1 but for the Lc in the 700- to 500- and the 500- to 240- mb layers. Heating rate units are"C/daY.

choice of particular HIRS channels depend upon the altitudeof the surface and the cloud layer in question. This form ofthe equation led to nvs errors of less than 10 Wm'z for clearand overcast conditions using the Phillips et o/. (1988) dataas well as for the global, one-day, clear-sky data set notedabove, The Lee technique is in the process of being recodedfor use in the operational technique, and the details concern-ing this technique will be submitted for publication in thenehr future. A similar analysis is underway for the clear-skyLc to account for the underestimation at high water vaporamounts.

Quasi'0perational Data Collection, Analysis' andDistributionThe clear-sky algorithms for the LC and oLR were imple-mented as an experimental quasi-operational system for fur-ther evaluation in May 1992, and the availability of the datafor use and evaluation by the international community wasannounced by letter and OMNET bulletin board in fanuary1993, The operations proceed as follows.

The daily radiance observations by the Tovs HIRS/2 in-struments onboard the NoAA-11 and Noaa-rz polar orbitingsatellites are operationally processed by Nona to producetheir five-day rotating sounding product archive (DsD3). Asubset of the DSDs is extracted by NOA{NEsDIS and modified

PE&RS

with our equations to include the clear'sky oLR and Lc forfour atmospheric layers. Each observation is considered rep-resentative-of the quantity associated with an area coveredbv a 3 by 3 array of individual HIRS scan spots [an area ofa6out zso km, it nadir). This high resolution data set is pro-duced at NESDIS and then transferred to ult{D via the Internetfor analysis and further processing.

At uvnlcrcs, 2.5' Iatitude by 2.s'longitude griddedfields are produced for each satellite and the four orbit dailymeans are used to enhance quality control and analysis ofthe data. Global images and bulk statistics are-produced toquickly assess data integrity and to aid in analysis utilizingdata visualization products,

The data sets iunently available include the high resolu-tion fields (as they are obtiined directly from NEsqIs) and the2,5" by z.sd gridd6d fields for both satellites' The daily aver--aee, o-btained from both satellites, is also included at the 2.5"bi z.s" resolution. As the data are collected, special tempo-riily a.reraged data are produced, including-five-day- andmoithly mian 2.5' by i.s'gridded fields. The montlly meandata ar6 also available as contoured images and can beviewed as a time series loop. The current archive includesdaily observations from lune 1992 to tlre present.-In

order to facilitate the transfer of data from the Univer-sitv of Marvland to the site of an interested user, an X-Win-dow based, menu-driven, interactive database facility is

313

-

3020

Flux Differences .r' i'-_ -_ -_ -_ lr -__ i__-

MRF. HIRS i i

7ll;;: .rE

E

100

-10-20-30 L

90 45 0 -45Latitude (Degrees)

-90

Figure 2. As in Figure 1 but forthe oLR and DLR.

available via a standard rCp/tp telnet session. A complete de-scription of the data, including naming conventions and fieldformats, is available via on-line documentation. Interestedusers may obtain the necessary login information through e-mail contact with dave@atmos.umd.edu.

A list of the organizations which have successfullytransferred data include

r Bureau of Meteorology Research Center, Melbourne, Aus-tralia;

. Department of Oceanography, Dalhousie University;o Institute for Physics, GKSS Forschungszenstrum, Geestchacht,

Germany;o fet Propulsion Laboratory, Pasadena California; and. NoAAAIMFS, Narragansett, Rbode Island.

It should be noted that the om fields will be added and ana-Iyzed from the beginning of our archival period when thecoefficients for the Lee (1993) model are derived for theNoAA-11 and uoaa-rz HIRS insbuments. It should also benoted that cloud effects may be added easily when cloudanalyses for the period become available from either theISCCP or the NESDIS procedure now under development. Fur-thermore, we plan to add analyses of the total oLR from theHIRS observed (raw) radiances during 1993. This will allowan operational determination of the top of the atmosphere"cloud forcing" being studied by a variety of climate groups.

ProducVTechnique Evaluation within ttlOAAThe LC, OLR, and DLR estimation techniques are experimen-tal, and their adaptation for operational implementation mustundergo a thorough review. The process for introducing anew product into operational production within NESDIS in-volves interactions with several standing NESDIS groups, andthe t c and DLR inference techniques are still in the evalua-tion process, The groups and their functions are describedbelow along with the status of the inference schemes relativeto them.

314

Research and Development Council (RDC)The Research and Development Council is a forum for thereview and evaluation of research and development projectsin NESDIS. Typically, new research ideas are surfaced at theRaD Council plenary sessions. Plenary sessions are open toall personnel within and outside of xnsors, so that discus-sion of research activities benefit from diverse points ofview, generally resulting in improved research plans. Re-views of the research during various developmental and test-ing phases may also be presented at an R&D Council plenarysession. The concept of the t-c inference scheme was dis-cussed with the RDb before the start of the project, and theresults of our preliminary results were described to them inthe fall of tggt,

Satellite Product Review Board (SPRB)The Satellite Product Review Board (senn) reviews all newproduct efforts at the times they are ready to move from de-velopment phase to the operational test phase and from theoperational test phase to the operations phase. Developmentphase is a design of the procedure to generate a data productfrom satellite radiances; it includes initial testing by the de-veloper, Operational test phase insures that the product willmeet efficiency and resource constraints in an operationalenvironment. it is also the time when the produit is pro-duced on an experimental basis, for a limiied time, and as-sessments of usefulness and expected reliabilitv as anoperational product are made. 'ihe operational phase rep-resents the long-term routine production and distribution ofthe data product by NESDIS. New products must be approvedby the srnr before they can become operational.

This project has had an informal review at the SnRB be-fore the leginning of our current operations test. The proj-ect will be reviewed asain earlv in 1994 after the userihave had at least 12 nionths to use the data in an oDera-tional setting.

Product (hersight Panels (P0P)Product Oversight Panels were established at NrSptS to pro-vide end-to-end scientific and technical oversight and recom-mendations to the SPRB regarding proposed new products,and improvements, modifications, and deletions of existingproducts, The PoPs are composed of scientists from the urS-DIs Office of Research and Applications, and Office of Satel-Iite Data Processing and Distribution. Outside scientists maybe invited to participate in the Pop. The poP recommended'the project for continued development in the fall of 1991,

Summary, Gonclusions, and Suggestions lor FutureActivitiesRadiative transfer theory and regression analysis have beenused to develop a technique to infer vertical profiles of long-wave radiative cooling in the atmosphere from radiance da[aobserved by the HIRS instrument flown on NOAA operationalsatellites. The analysis shows that cooling rates in the layersfrom the surface to 700 mb, 700 to 800 mb, 800 to 240 mb,and 240 to 10 mb may be estimated to within about i 15percent Rtr,ts for individual radiance observation sets. Theseerrors should decrease as the length of the averaging periodincreases,

An initial test of the technique in an operational frame-work was performed by collecting data from the NOAA-1o sat-ellite during a one-month period between 15 December 1990

and 15 January 1991. Clear-sky global fields of oLR, DLR, andLc were obtained on a 2,5o by 2,5'latitude-longitude grid,and monthly averages were produced and compared withcalculations made by the NMC trlnr', The agreement was gen-erally good, especially for zonally averaged quantities. Therewere, however, significant differences over many continentalareas, particularly Africa and Australia. A detailed analysisof these differences led to an improved technique for esti-mating the DLR, and it provided insights to NMC concerningneeded improvements in the MRF techniques.

The clear-sky algorithms for the Lc and oLR were imple-mented as an experimental quasi-operational system for fur-ther evaluation in May 1992, and the availability of the datawas announced to the international climate community inJanuary 1993. The data from the technique are currentlybeing used by several climate and medium range weatherforecast groups. Most groups have only had access to thedata for a few months, and this is too short a time to assessits usefulness in their operations.

However, from the perspective of utilizing a preciselydetermined longwave cooling rate from space observationwithin the operational analysis/forecast system, two positivearguments arise. The first is as an evaluation mechanism forthe current within-model procedures. The second is that, ifsuccessful, the determinations can be input into the model asa type of forcing function. In two recent studies, Kahn ef a/.(1994) utilized one year of wptc data (1985-86) to investigatethe meridional energy transports determined from the opera-tional analyses, and Yang ef o/. (1994), in a prognostic study,examined the column cooling distribution in medium rangeforecasts from the twc global model. Both studies demon-strated that the NMC forecast system in effect at that timegenerated excessive longwave radiative cooling in the trop-ics. As a consequence, the Hadley cell and walker circula-tions were weakened, and the energy divergence in thetropics was over 35 Wmz less than derived from net radia-tion measurements by ERBE. It is possible, then, that success-ful, operationally derived data from the HIRS instrumentscould have an impact in improving the operational analysesand forecasts.

There are additional applications of the data that we areaware of or that are under study. For example, NOAA/I{MChas archived the global distributions of clear-sky radiationmodel calculations using MRF initialized fields for about halfof our archival period, and they plan to compare many oftheir fields to ours, thereby providing an evaluation of theradiation physics in the tvtps as well as testing the initializa-tion schemes. As the NEDSIS operational cloud products be-come available, these comparisons and our data will gainadded significance as they will lead to the initialization ofthe forecast models with a better estimate of the vertical dis-tribution of radiative cooling. Similarly, information on themonthly and seasonal changes in the large-scale surface en-ergy budget and atmospheric cooling fields will likely re-olace indirect inferences now made from the oLR alone andused in climate diagnostics and forecasts by the Noaa Cli-mate Analysis Center. Furthermore, the improved DLR radia-tion scheme developed as an outgrowth of this project, andthe experimental operational data collection now underway,will lead to a more complete description of the global surfaceenergy budget, a term poorly understood, and one under ex-tensive study by the World Climate Research Program.

This project has proceeded relatively smoothly and hasaccomplished most of its goals, but some of the problems

PE&RS

that arose might have been alleviated by better informationat the time of the init ial planning stage. The major problemareas concern a determination of the necessary computerresources, ease of access of data from federal agencies,interagency agreements concerning continuation of fundingduring budget crises, and obtaining feedback from users ofthe new product. A determination of the required com-puter resources is difficult unless one limits the analysesat the outset. However, as there tends to be a naturalgrowth in curiosity and computer power as projects pro-ceed, it is perhaps best to extend ones init ial estimate ofneeds or to allow expansion of the computer resources asthe project proceeds.

Our access to data from NoAa was exceptionally good.However, it was sometimes impaired by operational con-straints or prior commitments. It is quite likely that this oc-curs with every agency, and investigators should attempt tomake arrangements for data access well in advance of whenthe data are actually required.

One of the major factors decreasing the pace of the proj-ect was the aperiodic funding of the program, Although fullfunding of the program was anticipated from the outset, va-garies in the political budgetary process often delayed the re-ceipt of funds beyond the anticipated date. In the absence ofspecial agreements or financing, such activities require oneto simultaneously maintain more than one project in order toinsure the continued funding of required personnel. This isnot always possible or desirable, particularly in a universitysetting where much of the research is being done by degreeseeking students.

Perhaps our biggest disappointment to date concernsfeedback from users of the product. However, it should bekept in mind that there are many new projects underway inthe international climate community, and most investigatorshave precious little time to use new data that may be outsideof the immediate scope of their projects or which may be en-tirely new. Our experience with other projects suggests thatpublished results are the best drawing card. This, however,takes a considerable amount of lead time. Alternatively,workshops to examine special problems are usually verystimulating. Such activities require planning at the beginningof the project as well as agency support.

Although it is too early to evaluate the usefulness ofthese data to the various institutions, our experiences withIrtoAA lead us to believe that the continued use of these datawill lead to better climate analyses and data assimilationtechniques of the radiation fields for use in forecast models,The technique is under continuing evaluation by NoAA forpossible operational implementation. In the interim, it iscontinuing under the sponsorship of Noaa's Climate andGlobal Change Program in the Operational Measurementssection, and we will report on the data and its use in futurepublications-

Finally, the estimation technique would likely be im-proved and be more easily extended to future satellite instru-ments if the radiation calculations were performed with aline-by-line model so as to more accurately account for all ofthe physics. Such detailed calculations were a monumentaltask at the start of this project because of the large amount ofcomputer time required for the calculations. Furthermore, itwas difficult to justify such calculations on the basis of theIcRcCM results and the uncertainties associated with estimat-ing cloud effects. Nevertheless, such calculations are moreeasily accomplished now and they will be necessary if the

technique is to be applied to higher spectral resolution in-struments, such as the Advanced Infrared Sounder (ans),planned for implementation in the late 1990s.

AcknowledgmentsThis research was sponsored in part by the National Aero-nautics and Space Administration through Grant NAGW-1477,54 and by the National Oceanic and AtmosphericAdministration through the Cooperative Institute for ClimateStudies.

ReferencesAnderson, R. L., and T. A. Bancroft, 1952. Stotrstrcal Theory in Re-

seorch, McGraw Hill Book Company, Inc., New York, 3s9 p.Barkstrom, B. R., E. Harrison, G. Smith, R. Green, J. Kibler, R. Crest,

and the ERBE Science Team, 1989. Earth Radiation Budget Ex-periment (ERBE) archival and April 1985 results, BulL Amer.Meteo r. S oc., 7 01725 4-7262,

Burch, D., and R. Alt, 1984. Continuum Absorption in the 700-1200cm-1 and 2400-2800 cm-t Windows, Report AFGL-TR-S4-0128,46 p.

Cox, S. K., D. S. McDougal, D. A. Randall, and R. A. Schiffer, 1987.FIRE - The First ISCCP Regional Experiment, Bull. Amer. Me-teor. Soc., 68:114-118.

Darnell, W. L., S. K. Gupta, and W. Staylor, 1983. Downward long-wave radiation at the surface from satellite measurements, /. C,D-mate Appl. Meteor., 22:1956-1960.

DOE, 1990. Atmospheric Radiation Measurcment Program Plan,DOE/ER-0441, U.S. Department of Energy, Office of Energy Re-search, Washington, D.C., 116 p,

Efroymson, M. A., 1960. Multiple regression analysis, Chapter 7,Mathematical Methods for Digital Compufers, John Wiley andSons, New York, New York.

Ellingson, R. G., and K. Campana, 1984. Tests of radiative effects ona medium-range numerical weather prediction model, RecenfProblems in Atmospheric Radiation (G. Fiocco, editor), DEEPAKPublishing, Hampton, Virginia, pp. 82-84.

Ellingson, R, G., and J. C. Gille, 1978, An infrared radiative transfermodel, Part 1: Model description and comparison of observa-tions with calculations, l. Atmos. Scj., 35:523-545.

Ellingson, R. G., and G. Serafino, 1984. Observations and calcula-tions of aerosol heating over the Arabian Sea during MONEX, /.Atmos. Sci. , 41:575-589.

Ellingson, R. G., D. J. Yanuk, H-T. Lee, and A. Gruber, 1989a. Atechnique for estimating outgoing longwave radiation from HIRSradiance observations, J. Atmos. Ocean. Tech., 6t206-71I.

Eliingson, R. G., D. f. Yanuk, and A. Gruber, 19s9b. On the effects ofthe choice of meteorological data on a radiation model simula-tion of the NOAA technique for estimating outgoing longwaveradiation from satellite radiance observations, l. Climate, S:767-765.

Ellingson, R. G., t. Ellis, and S. Fels, 1991. The intercomparison ofradiation codes in climate models (ICRCCM): Longwave results,l. Geophys. fies., 96:8929-8953.

Ellingson, R. G., H-T. Lee, D. Yanuk, and A. Gruber, 1994. Valida-tion of a technique for estimating outgoing longwave radiationfrom HIRS radiance observations, J. Atmos. Ocean. Tech., inpress.

Fels, S. B., and M. D. Schwarzkopf, t975. The simplified exchangeapproximation: A new method for radiative transfer calcula-tions, /. Atmos. Sci., 32i7475-1488.

Frouin, R., C. Gautier, and J. ]. Morcrette, 1988, Downward longwaveirradiance at the ocean surface from satellite data: Methodologyand rn sjfu validation, J. Geophy. Res., 93:597-619.

Gruber, A., and J. S. Winston, 1978. Earth-atmosphere radiative heat-

ing based on NOAA scanning radiometer measurements, 8u/,1.Amer. Meteor. Soc., 59:1570-1573,

IMSL, 1984. User's Manual, IMSL Library, FORTMN Subroutrnesfor Mathematics ond Stafi'stics, IMSL, Houston, Texas.

Kann, D. M., S.-K. Yang, and A. f. Miller, 1994. Mean meridionaltransport of energy in the Earth-atmosphere ststem using NMCglobal analyses and ERBE radiation, to appear in Tellus.

Kneizys, F. X., E. P. Shettle, L. W. Abreu, f. H. Chetwynd, G. P. An-derson, W. O, Gallery, J. E. A. Selby, and S. A. Clough, tSea.Users Guide to LOWTRAN Z AFGL-TR-8S -0t77 , Ait Force Geo-physics Laboratory, Hanscom AFB, Massachusetts, 1S7 p.

Lee, H-T., \993. Development of a statistical technique for estimat-ing the downward longwave rcdiation at the surface from satel-lite obsenations, Ph.D. Dissertation, Universitv of MarvlandCollege Park, 150 p.

Lee, H-T., and R. G. Ellingson, 1990. A regression technique for esti-mating downward longwave flux at the surface from HIRS radi-ance observations, Proceedings of the Seventh Conference onAtmospheric Radiation of the Anterican Meteorological Society,pp .2s7-252.

Liou, K. N., and Y. Xue, 1986. Exploration of the remote sounding ofinfrared cooling rates due to water vapot, Proceedings of theSixth Conference on Atmospheric Radiation of the AntericanMeteorologr Society, pp. 132-134.

Malkmus, R,, 1967. Random Lorentz band model with exponential-tailed S-1 line intensity distribution function, l. Opt. Soc. Amer.,57:323-329.

Phillips, N , f. Susskind, and L. McMillin, 1988. Results of a jointNOA,\AIASA sounder simulation study,,l. Atmos. Ocean Tech.,5:44-56.

Roberts, R. E., L. M. Biberman, and f. E. A. Selby, 1976. Infraredcontinuum absorption by atmospheric water vapor in the 8-10pm window. I. Appl. Opt., ts,2085-2090.

Rodgers, C. D., and C. D. Walshaw, 1966. The computation of in-frared cooling rates in planetary atmospheros, Quail. J, Roy. Me-teon Soc., 92:,67-92.

Schmetz, P., 1984. Grossraumige Bestimmung der Gegenstrahlungaus Satellitenun Analysendato, M.S. Thesis, Universitat Kdln,9 / D .

Schwarzkopf, M. D., and S. B, Fels, 1991. The simplified exchangeapproximation revisited: An accurate, rapid method for caompu-tastion of infrared cooling rates and fluxes,,I. Geophys. Res.,96:9075-9096.

Shaffer, W. A., and R. G. Ellingson, 1990. A technique for estimatinglongwave radiative cooling rates from HIRS radiance observa-tions, Proceedrngs of the Seventh Conference on AtmosphericRadiation of the Arnerican Meteorological Society, pp. 182-183.

Slingo, J. M., and A. Slingo, 1988. The response of a general circula-tion model to cloud longwave radiative forcing. I: Introductionand initial experiments, Quart. J. Roy. Meteor. Soc., L!4:7027-t062.

1991, The response of a general circulation model tocloud longwave radiative forcing. II: Further studies, Quart. J.Roy. Meteon Soc., 7L7i333-364.

SPLUS, 7992. Statistical Models in S, (J. M. Chambers and T. J. Has-tie, editors), Wadsworth & Brooks/Cole Advanced Books & Soft-ware, Pacific Grove, California,

WCP-115, 1986. Report of the Workhop on Surface RadiationBudget for Climate Applications (J. T. Suttles and G. Ohring, ed-itors), WMO/TD - No. 109, 144 p.

Yang, S.-K., W. Ebisuzaki, R. f. Lin, D. Kann, K. A. Campana, and A.f, Miller, 1994, Divergence of atmospheric energy flux and cloudradiative forcing from NMC medium range model forecasts andAMIP simulation for October 1986, Proceedings of the Bth Con-ference on Atmospheric Radiation of the American Meteorologi-cal Society, AMS, Boston, Massachusetts.

PE&RS